diff --git a/.doc/audit/ISSUES.md b/.doc/audit/ISSUES.md index 53ded071..fd05c662 100644 --- a/.doc/audit/ISSUES.md +++ b/.doc/audit/ISSUES.md @@ -6,6 +6,11 @@ | issue_id | 模块 / 检索词 | 问题 | 严重度 | issue_state | owner | 证据日期 / 范围 | 唯一主记录 | 经验 | |---|---|---|---|---|---|---|---|---| +| ISSUE-20260928-task-write-verb | intent / hasTaskWriteVerb / 帮我定 / 制定 / AMBIGUOUS_FALLBACK | 「帮我定/制定」设计开发计划任务书未进入任务触发,对话只有掂量阶段模板 | medium | resolved | Cursor | 2026-09-28;test sess-p8-u33 msg 89/90;意图单测 | [主记录](../ops/short-term/2026-09-28-test-user-doc-issues.md) | [LESSON-20260928-task-write-verb-visible-refusal](../ops/lessons/2026-09-28-task-write-verb-visible-refusal.md) | +| ISSUE-20260928-workflow-scope-reply | coordinator / ClarificationQuestion / ErrPrerequisiteNotMet / rd_design_plan | 跨工作流澄清与研发前置未满足时,原因没有进入对话 | medium | resolved | Cursor | 2026-09-28;协调器单测。test 原例停在动词未命中,未走到本分支 | [主记录](../ops/short-term/2026-09-28-test-user-doc-issues.md) | [LESSON-20260928-task-write-verb-visible-refusal](../ops/lessons/2026-09-28-task-write-verb-visible-refusal.md) | +| ISSUE-20260928-quality-failed-success-copy | bizdev / quality_failed / 正文已保存至 / output.md | 质量未通过仍声称正文已保存,并写出工作区绝对路径 | medium | resolved | Cursor | 2026-09-28;test cdo_quote_1790560155851280170;同步回复单测 | [主记录](../ops/short-term/2026-09-28-test-user-doc-issues.md) | [LESSON-20260928-task-write-verb-visible-refusal](../ops/lessons/2026-09-28-task-write-verb-visible-refusal.md) | +| ISSUE-20260928-pi-runner-data-dir | agent / PI runner / DATA_DIR / MISSING_TARGET / 18866 | test 启动时根 .env 的 dev runner 地址盖过本环境 sidecar,工作区被重写到空目录 | high | resolved | Cursor | 2026-09-28;dev `4f581306`。当时 test 二进制仍是修复前制品 | [验收报告](../test-reports/2026-09-28-pi-runner-env-mismatch-test-report.md) | — | +| ISSUE-20260928-material-chunk-embed | knowledge / chunk index failed / 8192 / 34635 tokens | 材料分块批超过嵌入模型上下文,索引失败 | medium | open | — | 2026-09-28;test 日志 project 8,文件 W020251117353305579317.docx | [主记录](../ops/short-term/2026-09-28-test-user-doc-issues.md) | 本轮不修 | | ISSUE-20260928-attachment-only-turn | chat / chat/stream / message cannot be empty / HTTP_400 / attachments | 只上传文档、不输入文字时 turn 被空消息 400 拒绝,前端不读信封,用户看不到任何回复 | high | resolved | Cursor | 2026-09-28;dev :8080 重载后浏览器仅附件得到梳理审阅,无关文字先回答再追问;pending 分支 API 200 | [主记录](../ops/short-term/2026-09-28-attachment-only-turn.md) | 空消息门禁要放在附件判定之后;4xx 必须读信封。见[测试报告](../test-reports/2026-09-28-attachment-only-turn-test-report.md) | | ISSUE-20260928-task-run-replay-502 | web / useRecoveredTaskPhases / tasks/runs / ERR_HTTP2_PROTOCOL_ERROR / 502 | 历史失败气泡恢复订阅 task run 事件时出现 HTTP/2 协议错误和 502,与附件空消息 400 无因果关系 | medium | investigating | Cursor | 2026-09-28;用户控制台 `run_1790558010622744831`;本轮未改订阅逻辑 | [主记录](../ops/short-term/2026-09-28-attachment-only-turn.md#范围外) | 未修,保持观察 | | ISSUE-20260928-activation-preview-precheck | web / projectSettings / activation-preview / NO_GOVERNED_DOCUMENT / clientDutyIssues | 全部文档未纳入人工审批时前端仍 POST activation-preview,控制台 400 | medium | resolved | Cursor | 2026-09-28;dev 项目 6023 默认计划无 activation-preview 请求,兼任后 preview 200 | [主记录](../ops/short-term/2026-09-28-activation-preview-precheck.md#issue-20260928-activation-preview-precheck) | [LESSON-20260928-activation-preview-client-gate](../ops/lessons/2026-09-28-activation-preview-client-gate.md) | diff --git a/.doc/ops/lessons/2026-09-28-task-write-verb-visible-refusal.md b/.doc/ops/lessons/2026-09-28-task-write-verb-visible-refusal.md new file mode 100644 index 00000000..13ce0a33 --- /dev/null +++ b/.doc/ops/lessons/2026-09-28-task-write-verb-visible-refusal.md @@ -0,0 +1,20 @@ +--- +lesson_id: LESSON-20260928-task-write-verb-visible-refusal +lesson_state: candidate +owner: Cursor +reviewer: Cursor(自审,待含本提交的运行制品复验原句) +last_reviewed: 2026-09-28 +reviewed_revision: db859e185ebcd0e7eccd2241ca07f0e9f4fe4c6b +source_issues: [ISSUE-20260928-task-write-verb, ISSUE-20260928-workflow-scope-reply, ISSUE-20260928-quality-failed-success-copy] +keywords: [intent, hasTaskWriteVerb, 帮我定, ClarificationQuestion, ErrPrerequisiteNotMet, quality_failed] +--- + +# 任务写动词要覆盖「定」,拒绝原因要进对话 + +- 触发条件:用户说「帮我定/制定/拟定」某交付物,对话却进入「再掂量一下」;或任务被工作流范围、研发前置、质量门禁挡住,气泡仍像成功或只有协调失败。 +- 机制与来源:强写动词表原先只有生成/做/跑/出/编写/编制/起草。「定」单独出现会误伤「决定」,所以只认「制定」「拟定」以及「帮我/请/给我/帮忙」后面的「定/拟」。跨流澄清和前置失败必须成为用户可见回复,不能再被阶段模板或错误事件盖掉。质量未通过不要写宿主机路径。见 ISSUE-20260928-task-write-verb。 +- 开发检查:新增交付动词时补强动词或祈使短语,并加一条反例(决定、看看)。`RequiresClarifying` 的问题原文要到达会话。`ErrPrerequisiteNotMet` 的中文原因用拒绝态返回,错误值为空,避免会话层丢掉 `NaturalReply`。 +- 验证方式:`go test ./internal/intent/ ./internal/coordinator/` 中写动词、跨流澄清、前置可见、质量失败文案四类用例。原句在运行制品上的复验完成前,本条保持 candidate。 +- 适用边界:只覆盖中文任务触发与同步回复。材料嵌入分块超限、PI runner 数据目录错配是相邻但独立的问题。 +- 防护落实:`doc_workflow_scope_test.go` 与 `pending_confirmation_reply_test.go`。 +- 修订与失效:若产品明确「定」永不表示编写,应删祈使短语并退役本条。 diff --git a/.doc/ops/lessons/INDEX.md b/.doc/ops/lessons/INDEX.md index 7f4eb686..12baec2f 100644 --- a/.doc/ops/lessons/INDEX.md +++ b/.doc/ops/lessons/INDEX.md @@ -4,6 +4,7 @@ | lesson_id | 触发条件 / 关键词 | 注意事项 | lesson_state | 主记录 / 最近复核 | |---|---|---|---|---| +| LESSON-20260928-task-write-verb-visible-refusal | 帮我定、制定、ClarificationQuestion、前置条件、quality_failed | 写动词覆盖制定与帮我定;拒绝和澄清进对话;质量失败不写宿主机路径 | candidate | [写动词与可见拒绝](2026-09-28-task-write-verb-visible-refusal.md);2026-09-28 | | LESSON-20260928-activation-preview-client-gate | activation-preview、NO_GOVERNED_DOCUMENT、clientDutyIssues、400 | 服务端激活门禁要有同等前端预检;live_edit 不套激活门禁 | validated | [激活预检对齐](2026-09-28-activation-preview-client-gate.md);2026-09-28 | | LESSON-20260928-squad-commit-as-author | 小队落盘、uow permission denied、AutoCommitUserID、OnBehalfOf、非编制人 | 自动提交身份用文档编制人,触发者只写入 OnBehalfOf;不要放宽 ActionCommit | candidate | [编制人代理提交](2026-09-28-squad-commit-as-author.md);2026-09-28 | | LESSON-20260928-standard-role-reseed | 注册、role-guest、system_roles、is_system、23503、23505、INTERNAL | 硬编码标准角色须幂等补种并标 is_system;用户名冲突返回领域 400,不要包进 insert user | validated | [标准角色可自愈](2026-09-28-standard-role-reseed.md);2026-09-28 | diff --git a/.doc/ops/short-term/2026-09-28-test-user-doc-issues.md b/.doc/ops/short-term/2026-09-28-test-user-doc-issues.md index 98777728..66e1a9d6 100644 --- a/.doc/ops/short-term/2026-09-28-test-user-doc-issues.md +++ b/.doc/ops/short-term/2026-09-28-test-user-doc-issues.md @@ -32,7 +32,7 @@ remote_sync: not_requested - 检索词:`帮我定设计开发计划任务书`、`hasTaskWriteVerb`、`质量检查未通过`、`PI runner data_dir`。 - P1(只上传附件被空消息拒绝)已由 ISSUE-20260928-attachment-only-turn 在 dev 修复。 - catalog overlap(无 scope 绑定)hard/soft 为空。带 `--intent` 注册被 `same_scope` 拒绝,持有者 `agent-reply-llm-narrative-2026-09-28`(scopes `internal`+`.doc`)。本任务不改其文件(`dialogue_llm.go` / `intent_reply_llm.go` 等),只改意图动词与任务回复文案。 -- P3/P4 的 runner 数据目录错配由并行任务 `pi-runner-env-mismatch-2026-09-28`(`4f581306`)修复,合入时其 integrate 队列为 failed(DIRTY_TREE),本任务不代合入。 +- P3/P4 的 runner 数据目录错配由并行任务 `pi-runner-env-mismatch-2026-09-28`(`4f581306`)修复。开工时其 integrate 为 DIRTY_TREE;收口前已确认该提交是 local dev `193a752f` 的祖先。本任务不代合入、不改其文件。 ## 事实 @@ -48,3 +48,7 @@ test 库 `agent_system_test`(127.0.0.1:5434)会话 `sess-p8-u33`(2026-09-2 - 质量未通过或未完成时,同步回复改用 `CompletionSummary`,不展示宿主机路径、不声称正文已保存。 - `ErrPrerequisiteNotMet` 以中文前置条件回复结束回合,不再把哨兵错误当成协调失败丢掉回复。 - 模糊咨询若已带澄清问题,直接透传,不再覆盖成阶段对照模板。 + +## 验证 + +单测与 test 进程核对见 [测试报告](../../test-reports/2026-09-28-test-user-doc-issues-test-report.md)。报价单未在本轮重跑。test 进程 pid 1450126 已指向 18866 与 `data/test`,二进制来自主目录 dev `193a752f`,不含本提交。 diff --git a/.doc/test-reports/2026-09-28-test-user-doc-issues-test-report.md b/.doc/test-reports/2026-09-28-test-user-doc-issues-test-report.md new file mode 100644 index 00000000..331cfa6c --- /dev/null +++ b/.doc/test-reports/2026-09-28-test-user-doc-issues-test-report.md @@ -0,0 +1,96 @@ +# 第三方 test 用户文档问题 测试报告 + +## 一、 测试背景 (Background) + +第三方用户在 test 环境整理了《AI 测试问题》。只上传附件被空消息拒绝已由 ISSUE-20260928-attachment-only-turn 修复。本轮处理另外两类: + +- 「帮我定设计开发计划任务书」没有进入任务,对话是掂量阶段模板。 +- 「帮我生成报价单」气泡写成正文已保存到宿主机路径,工作区实际是质量未通过的阻塞说明。根因是 PI runner 数据目录错配,已由 `4f581306` 合入 local dev。 + +主记录:[2026-09-28-test-user-doc-issues](../ops/short-term/2026-09-28-test-user-doc-issues.md)。 + +## 二、 测试目标 (Test Objectives) + +- 「帮我定设计开发计划任务书」在商务流激活时进入任务,并带出研发流澄清;「决定」不触发编写。 +- 研发前置未满足时,用户看到中文原因,而不是协调失败。 +- 质量未通过的同步回复使用质量摘要,不写宿主机路径,不声称正文已保存。 +- 核对 test 进程是否已经指向本环境 runner 与 `data/test`。 + +## 三、 测试方案与环境 (Test Plan & Environment) + +- 源码:worktree `.worktrees/test-user-doc-issues-2026-09-28`,提交 `db859e185ebcd0e7eccd2241ca07f0e9f4fe4c6b`,基线 `8f77053f`。 +- 命令:`go test ./internal/intent/ ./internal/coordinator/ -count=1 -timeout 180s`,以及 `-run` 聚焦用例;`go vet` 同两包。 +- test 运行制品:`bin/agent-server-test` 修改时间 2026-09-28 04:04:03 +0000,主目录 `HEAD` `193a752f`(含 `4f581306`)。进程 pid `1450126`,`ENV=test`,`DATA_DIR=.../data/test`,`PI_AGENT_RUNNER_URL=http://127.0.0.1:18866`。18866 的 `data_dir` 为 `data/test/workspaces`。本提交的回复文案尚未打进该二进制。 +- 未用第三方账号再跑「生成报价单」。该操作会写入其项目;本会话没有登录凭证。runner 数据目录的产品链路证据见 [PI runner 环境错配报告](./2026-09-28-pi-runner-env-mismatch-test-report.md) 的 `bd_research` 实测。竞品视频未发现独立处理缺陷,未扩测。 + +## 四、 预期结果 (Expected Results) + +- 「帮我定设计开发计划任务书」+ 激活 `main`:`TRIGGER_TASK`,工作流范围不匹配,澄清含研发设计控制流程,`RequiresClarifying`。 +- 激活 `rdev` 时「帮我制定…」命中 `rd_design_plan` 且不拦截。 +- 「我决定先放一放,再看看进展」不是 `TRIGGER_TASK`。 +- 质量失败回复不含「正文已保存」和 `output.md` 绝对路径。 +- 前置未满足回复含「请先完成商务对接流程」,错误值为空。 +- 已完成回复仍含「正文已保存。」 +- test 进程 runner 为 18866,数据目录为 test。 + +## 五、 实际测试结果与判定 (Actual Results & Evaluation) + +### 用例 1:帮我定 + 商务流 + +- **输入**:`TestWriteVerb_DingDesignPlanCrossFlow` +- **实际输出**:`intent=TRIGGER_TASK branch=NEW_INTENT_PREEMPT confidence=0.86`,`PASS` +- **判定**:PASS +- **耗时**:包内 <0.01s + +### 用例 2:制定 + 研发流 + +- **输入**:`TestWriteVerb_ZhidingDesignPlanOnRdev` +- **实际输出**:同上 `TRIGGER_TASK`,`PASS`(断言任务键 `rd_design_plan` 且无范围拦截) +- **判定**:PASS + +### 用例 3:决定不是编写 + +- **输入**:`TestWriteVerb_DecideIsNotWrite` +- **实际输出**:`QUERY_PROGRESS` / `CASUAL_CHAT`,无 `TRIGGER_TASK`,`PASS` +- **判定**:PASS + +### 用例 4:质量失败文案 + +- **输入**:`TestFormatTriggerTaskReply_QualityFailedOmitsHostPath` +- **实际输出**:`PASS`(回复走质量摘要,不含宿主机路径) +- **判定**:PASS + +### 用例 5:前置条件可见 + +- **输入**:`TestSyncTriggerTask_PrerequisiteIsUserVisible` +- **实际输出**:`PASS` +- **判定**:PASS + +### 用例 6:成功文案仍保留「正文已保存。」 + +- **输入**:`TestSyncTriggerTask_Committed_KeepsSavedWording` +- **实际输出**:`PASS` +- **判定**:PASS + +### 用例 7:test 进程 runner 指向 + +- **输入**:读 pid `1450126` 的 `ENV`、`DATA_DIR`、`PI_AGENT_RUNNER_URL`;`GET` 18866 与 8081 `/health`;在二进制中查找 `RUNNER_MISMATCH` 用户文案。 +- **实际输出**:`ENV=test`;`DATA_DIR=/home/weichen/projects/Agent-System/./data/test`;`PI_AGENT_RUNNER_URL=http://127.0.0.1:18866`;18866 `data_dir=/home/weichen/projects/Agent-System/data/test/workspaces`;health `schema_version=78` `agent_backend=pi-agent`;二进制含「模型执行环境与数据目录不一致」。 +- **判定**:PASS(runner 接线)。报价单全链路本轮未运行。 +- **耗时**:健康检查 <1s。 + +日志:`.doc/test-reports/assets/2026-09-28-test-user-doc-issues/go-test-focused.txt`、`go-test-packages.txt`、`go-vet.txt`。 + +## 六、 自动化测试套件汇总 (Test Suite Execution) + +| 命令 | 结果 | +|---|---| +| `go test ./internal/intent/ ./internal/coordinator/ -count=1 -timeout 180s` | PASS(intent 0.078s,coordinator 0.015s) | +| 聚焦 `-run 'TestWriteVerb_\|TestFormatTriggerTaskReply_QualityFailed\|TestSyncTriggerTask_Prerequisite\|TestSyncTriggerTask_Committed_KeepsSavedWording'` | PASS | +| `go vet ./internal/intent/ ./internal/coordinator/` | 无输出,退出码 0 | + +未跑全量 `go test ./...`,也未跑 nightly 五项。本变更不改运行模板。材料嵌入超限(约 34635 tokens,模型上限 8192)只登记 ISSUE-20260928-material-chunk-embed,未测修复。 + +## 七、 测试结论与后续优化建议 (Conclusion & Next Steps) + +意图动词、澄清透传、前置可见、质量失败文案由单测覆盖,达到本任务的代码验收。test 当前制品已指向本环境 runner,runner 错配修复在 dev `193a752f`。本提交的回复文案还没进 `bin/agent-server-test`。报价单没有在本轮用第三方项目重跑。嵌入分块超限仍开放。 diff --git a/.doc/test-reports/assets/2026-09-28-test-user-doc-issues/go-test-focused.txt b/.doc/test-reports/assets/2026-09-28-test-user-doc-issues/go-test-focused.txt new file mode 100644 index 00000000..01505695 --- /dev/null +++ b/.doc/test-reports/assets/2026-09-28-test-user-doc-issues/go-test-focused.txt @@ -0,0 +1,23 @@ +=== RUN TestWriteVerb_DingDesignPlanCrossFlow +2026/09/28 04:05:41 INFO intent_recognition intent=TRIGGER_TASK branch=NEW_INTENT_PREEMPT confidence=0.86 +--- PASS: TestWriteVerb_DingDesignPlanCrossFlow (0.00s) +=== RUN TestWriteVerb_ZhidingDesignPlanOnRdev +2026/09/28 04:05:41 INFO intent_recognition intent=TRIGGER_TASK branch=NEW_INTENT_PREEMPT confidence=0.86 +--- PASS: TestWriteVerb_ZhidingDesignPlanOnRdev (0.00s) +=== RUN TestWriteVerb_DecideIsNotWrite +2026/09/28 04:05:41 INFO intent_recognition intent=QUERY_PROGRESS branch=NEW_INTENT_PREEMPT confidence=0.9 +2026/09/28 04:05:41 INFO intent_recognition intent=CASUAL_CHAT branch=AMBIGUOUS_FALLBACK confidence=0.2 +2026/09/28 04:05:41 INFO intent_recognition intent=QUERY_PROGRESS branch=NEW_INTENT_PREEMPT confidence=0.9 +--- PASS: TestWriteVerb_DecideIsNotWrite (0.00s) +PASS +ok chat-agent-system/internal/intent 0.008s +=== RUN TestSyncTriggerTask_Committed_KeepsSavedWording +2026/09/28 04:05:41 INFO coordinator_route intent=TRIGGER_TASK branch=NEW_INTENT_PREEMPT confidence=0 channel=SQUAD_FIXED reason=标准任务/阶段推进/工作流运行/长任务续跑指令,进入小队受控执行 +--- PASS: TestSyncTriggerTask_Committed_KeepsSavedWording (0.00s) +=== RUN TestFormatTriggerTaskReply_QualityFailedOmitsHostPath +--- PASS: TestFormatTriggerTaskReply_QualityFailedOmitsHostPath (0.00s) +=== RUN TestSyncTriggerTask_PrerequisiteIsUserVisible +2026/09/28 04:05:41 INFO coordinator_route intent=TRIGGER_TASK branch=NEW_INTENT_PREEMPT confidence=0 channel=SQUAD_FIXED reason=标准任务/阶段推进/工作流运行/长任务续跑指令,进入小队受控执行 +--- PASS: TestSyncTriggerTask_PrerequisiteIsUserVisible (0.00s) +PASS +ok chat-agent-system/internal/coordinator 0.007s diff --git a/.doc/test-reports/assets/2026-09-28-test-user-doc-issues/go-test-packages.txt b/.doc/test-reports/assets/2026-09-28-test-user-doc-issues/go-test-packages.txt new file mode 100644 index 00000000..b5eca2df --- /dev/null +++ b/.doc/test-reports/assets/2026-09-28-test-user-doc-issues/go-test-packages.txt @@ -0,0 +1,2 @@ +ok chat-agent-system/internal/intent 0.078s +ok chat-agent-system/internal/coordinator 0.015s diff --git a/.doc/test-reports/assets/2026-09-28-test-user-doc-issues/go-vet.txt b/.doc/test-reports/assets/2026-09-28-test-user-doc-issues/go-vet.txt new file mode 100644 index 00000000..e69de29b